Three years on from ChatGPT’s arrival, it’s still not clear what AI is doing to the labour market

Inconveniently for those of us who keep insisting AI will change everything, AI is yet to change very much

We are past the event horizon; the takeoff has started. Humanity is close to building digital superintelligence, and at least so far it’s much less weird than it seems like it should be.
Sam Altman, The Gentle Singularity, 11/6/25

If you had to pick the defining acronym in business culture for 2025, you’d be hard-pressed to look past AI. But DEI and ESG would not be far behind on the shortlist… During Covid and in its aftermath, acute talent shortages drove managers to focus on retention — office perks and relaxed attitudes to remote work followed. Now, with generative AI advancing rapidly, that equation has changed. Companies are less fearful about losing marginal staff, betting that a mix of high-performing talent and AI adoption can plug the gap.
John McDuling, Capital Brief, 1/12/25

It’s almost a year to the day that I had my Kasparov moment.

Long story short, I needed to ghostwrite a biography* in a ridiculously short period of time and had little option but to get ChatGPT to handle as much of the grunt work as possible.

AI didn’t write the book for me, but while racing to bang out a 60,000-word-plus tome, I kept being taken aback by just how intelligent artificial intelligence was.

I ended up feeling like the Harvard maths professor in Good Will Hunting who tells Will, “Most days I wish I never met you. Because then I could sleep at night, and I wouldn’t... have to walk around with the knowledge that there’s someone like you out there.”

I imagine Kasparov felt a similar sense of incredulous despair when he was defeated by Deep Blue in 1997. I fear many more people will experience a similar sense of disorientating superfluousness once they realise a machine can do what they do, only much quicker and cheaper.

And, increasingly, much better.

My watch has (almost) ended
After I finished writing the book and returned to my Substack labours early this year, I resolved to focus on AI throughout 2025.

I did this for two reasons. First, the AI revolution seemed like the world’s biggest story. Second, I was a legacy media journalist once and have some understanding of what occurs when a game-changing technology collides with a venerable industry. Having taken the industry-obliterating technological disruption ride myself, it seemed only gentlemanly to give others fair warning of what was in store for them.

After six months of banging on about why the true believers were on the money and the sceptics were living in denial, I belatedly realised both camps were talking past each other.

An AI doomer would point to rising graduate unemployment as evidence AI was starting to collapse the labour market. An AI sceptic would point out that overall unemployment rates were still relatively low.

An AI enthusiast would reference an AI model acing a bar or medical licensing exam. An AI naysayer would point out the same model was also given to making shit up.

And so on.

Back in June, I decided it would be helpful to keep an eye on various metrics – unemployment and underemployment rates, wage and productivity growth, job listings – to determine the impact of AI on the labour market.

At year’s end, here’s what the scoreboard reveals.

Unemployment
While there has been no shortage of high-profile layoffs during the year, especially in the tech sector, the overall unemployment rate remains respectably sub-5% in Anglosphere nations.

That noted, I wouldn’t get too comfortable.

There is widespread anecdotal evidence that while most businesses haven’t yet started trimming their workforces, many have instituted tacit or explicit hiring freezes. There’s now solid evidence that making it difficult for younger workers to begin their careers.

The labour market is still some way off being loose, but it’s no longer drum-tight. The debate continues to rage over why unemployment rates are drifting upward. Many argue we’re just witnessing the effects of standard economic cycles rather than the start of a jobs apocalypse.

Underemployment
Given the number of Goodbye, $165,000 Tech Jobs. Student Coders Seek Work at Chipotle-style articles I’ve waded through over the past 12 months, I suspected the underemployment rate data would be a slam dunk.

No such luck. As with unemployment rates, underemployment rates have drifted upwards but not dramatically. Part of the issue is that underemployment refers to both working fewer hours than you want, as well as working a job you’re overqualified for.

I suspect that many people who once had a white-collar job, or who assumed they would secure one after graduating, are making ends meet through gig economy work. Back in May, my fellow Substacker Shawn K went viral with a post (The Great Displacement is Already Well Underway) detailing how he’d unsuccessfully applied for 750+ software engineer jobs while working as a DoorDash delivery driver.

Job ads
Job postings trended down in 2025. But they’ve been trending downwards since the post-pandemic hiring frenzy of 2022.

That noted, the drop-off in job ads may be a leading indicator of what’s to come, given there have been noticeably steep declines in ads for tech and white-collar jobs.

Wage growth
The data isn’t a slam dunk for either side of the argument. Wage growth in the US fell to around 4 per cent in 2025, down from 5-6 per cent during the Covid era, and barely keeping pace with inflation.

Productivity
It looks like US productivity growth will come in at an unremarkable 1.5–2 per cent. That’s not terrible, especially compared to recent history, but it’s not cause for excitement either.

Straws in the wind
To summarise, the data is open to different interpretations by sceptics and true believers. The former argue that labour markets tighten and loosen all the time and there’s nothing remarkable about how the labour market functioned in 2025. The latter claim the data is starting to show what many white-collar workers fear – AI is coming for their jobs.

I did mention a couple of less easily quantifiable metrics in my June post.

I wondered if we would start to see prominent educational institutions, or entire national education systems, reinvent themselves for a post-AI world. There’s some evidence that has begun to happen, though nothing especially dramatic or widespread has yet occurred.

I also argued that if AI were as transformative as its boosters insist it is, it would inevitably come to dominate the news and political agenda. But while AI remains a hot topic, it’s not even as hot as, say, housing, immigration or cost of living pressures.

Let’s reconvene in a year
Well, this post turned out to be more anti-climactic than expected.

At the risk of grasping at straws, I suspect we are seeing a rerun of Solow’s paradox, in which people are starting to use a powerful new technology, but the effects are taking a while to show up in the data.

Much the same thing happened in the 1980s, prompting the Nobel Prize-winning economist Robert Solow to observe, “You can see the computer age everywhere but in the productivity statistics.”

Then again, maybe I’m wrong and the dwindling band cynics still insisting AI is an overhyped nothingburger will ultimately be proven right. I imagine things will be much clearer at the conclusion of 2026, so I’ll do a follow-up post then.

* Still available at all good bookstores, if you’re looking for a Christmas gift for the man in your life.